Abstract: The bicycle following behavior was described firstly, and the parameters about the bicycle traffic were collected by the video experiments at the signalized intersection. A series of bicycle following models with 3 different BP neural network structures were built considering the influences of some critical parameters in the bicycle traffic, such as the distance between bicycles, the relative speed, the acceleration of the leading bicycle, the transversal interference, and the character of the bicycle rider, on the acceleration of the following bicycle. Based on the comparative analysis of the training error of the different BP neural networks, a bicycle following model was proposed. This model can better reflect the stimulusresponse relationship between the leading and the following bicycles in the bicycle colony at the signalized intersection. This is of reference for analyzing the microscopic characteristics of the mixed urban traffic and performing corresponding microscopic simulation.

Key words: engineering of communication and transportation safety, bicycle traffic, following behavior, bicyclefollowing, BP algorithm, signalized intersection